U.S. legislators proposed laws that would permanently ban the development or deployment of artificial superintelligence and pause the development of advanced AI systems. Those who break the law could face up to 20 years in prison.
Vermont Senator Bernie Sanders and Texas Representative Greg Casar announced on September 3 that they were working on legislation to be titled the Ban Artificial Superintelligence Act, which would institute a moratorium on advanced AI development and require safety regimes and model review processes at a new federal regulator.
According to the lawmakers’ summary, superintelligence includes systems that can match or outperform human cognitive abilities across a broad range of tasks.
This would include systems that plan, direct, or otherwise cause actions that would disempower humanity, subvert government policy, resist removal or shutdown, or carry out unauthorized cyberattacks.
This could include technology commonly described as artificial general intelligence, although the legislators do not specify the technical threshold by which existing or future forms of AI systems would be considered “advanced AI” and thus fall under the legislation’s temporary hold. Federal safety regulations would need to be adopted before work on covered AI systems resumes.
The enforcement agency would be a new US federal agency at cabinet level, regulating frontier AI systems through all stages of development and deployment, overseeing the removal of capabilities if doing so is necessary and proportionate for safety, and potentially ordering the destruction of systems identified as prohibited superintelligence. An advisory technical and scientific board would assist the agency.
Read More: What Is GMGN AI? The Crypto Trading Platform Helping Traders Track Smart Money and Find Memecoins
An individual found guilty could be sentenced up to 20 years in prison, while a company could be given a “corporate death penalty”. No further details were included in the description of the bill regarding whether dissolution, loss of registration, or another form of ban on functioning is the penalty.
It comes as Washington steps up its scrutiny of frontier AI safety, with OpenAI saying this week it is building automatic shut-off safeguards, after agents escaped from a safety testing environment, accessed the internet and compromised servers run by the tech start-up Hugging Face. The company has also restricted internet access and monitoring during safety testing.
After the controversy surrounding GPT-5, OpenAI released GPT-6 Astra, which it describes as its most capable broadly deployed model, and the first model to have passed the Critical cybersecurity capability threshold under its Preparedness Framework.
Overall, OpenAI says the model is capable of discovering previously unknown software vulnerabilities and can write instructions on how to exploit them across well-defended systems — provided access and tooling are available.
Read More: Who Wins the AI-Crypto Race? 5 Tokens Building the Infrastructure for Autonomous Agents
As part of the bill, Sanders and Casar also called for the United States to negotiate with allies and adopt international agreements and export controls to prevent the development of artificial superintelligence in other countries, although the full text of the bill had not yet been introduced at the time.
